mojuba

I think the number per active yearly users would be even more interesting. Clearly there is a correlation with the traffic too. But I wouldn't be very surprised if the relative numbers were still in decline.

Strictly speaking that's not holography if I understood it correctly. Holography is based on interference created by a surface with a special pattern, and it doesn't require any matter other than the diffractive surface. Whereas here they say they move "particles" (what particles? dust? atoms?). Still pretty cool of course and a thing on its own.

To be fair, LLMs can still derive those kinds of things from text, at the very least from your own comment if it made it to the training set though I'm sure it is mentioned in a lot of other places already. Many of this type of mistakes went away after reasoning was introduced.

But I'm sure you can still find tasks that they will have difficulty solving, involving the most fundamental concepts that can only be experienced in the physical world to be understood well, like left and right, near and far, hot and cold, heavy and light, etc.

Sorry but I don't think you read any of the linked articles. Just one example of an anti-pattern that is very common and is undeniably harmful: Bad Revenue.

It comes down to deceiving or annoying your customers. E.g. making cancellation difficult or confusing, tricking the users into buying yearly subscription by showing them the monthly price of the yearly subscription (explicitly prohibited by Apple in mobile apps btw) and many others which can boost your revenue short-term but do a lot more harm in the long term, or eventually even kill your company.

These sales anti-patterns can work for bigger companies that are long past their growth stage and are now looking for ways to squeeze pennies out of every customer. Early stage or growth stage startups should not copy these practices: growth is a very fragile thing that can be destroyed by negative word of mouth easily.

Generally, facts over instructions. This has been Anthropic's recommendation too, in one of their recent blog posts.

Also the shorter the better, let the model figure out the rest. Overinstruction degrades intelligence. We tend to underestimate their capabilities, we overinstruct them and then complain about them being dumb.

One unexpected discovery that I have made while building an AI-based system: the LLM's are bad at designing prompts.

We tend to think that the AI has some sort of self-knowledge and should be good at designing prompts for itself but it's really not.

Been struggling with a task that heavily depended on prompts, ended up rewriting all my prompts from scratch in my own words, and it finally worked. Then every time I ask Claude to fix something in the prompts, it invariably makes it worse.

A very strange phenomenon that can probably be explained by the quality of prompt design advice that made it to the training dataset. Bottomline, all the prompt design advice that you can find on the internet is really not great.

True, tried it so many times and every time I come up with something myself though sometimes inspired by the AI's ideas.

Verifying trademarks and domain name availability is usually an additional step you need to ask it to perform. Trademark DB searches by the way are intentionally made difficult to scrape so most of the time it's a manual process anyway.

However, once you give it all the information (TM search results, domain name availability) it can help you with the judgement of how safe the name is from the legal perspective. With the obvious caveats, but still a good starting point if you are serious about the name.

Are you measuring the number of bugs per line of code in production, compared to pre-AI? That's the metric.

> The result would be worse code, probably. Also probably still good enough.

And that's because the definition of "good enough" is drifting towards "crappy", has been for decades, now accelerated by the AI.

Was thinking the other day how the bars of quality for software pre-Internet was so much higher just because any small software shop only ever had one shot: ship your software on floppy disks, with practically no chance to patch and fix bugs later. People either like your 1.0 or they don't and that's your whole business story.

Then came the Internet, bug fix updates became so much easier, and with it came crappification and bugs-as-a-norm attitude.

And now this.

Not whining, just stating the facts.

This is all good to know, but funny how we are suddenly back to formal languages and commands. Aren't these things intelligent enough to figure these things out for us?

Exactly, and nothing will ever match the "prestige" of .com

.ai is also kind of cool right now but it will probably lose its appeal after some years when everything will become AI and the world will get tired of the term.

But .com is still at the top of the food chain. If you can survive for a few years, raise your series A/B, you will likely have enough money to acquire your sexy single-word .com name.

Agreed and that's probably the second major killer feature of C++.

Yes to all that but still, my pet theory is that languages rise on one or two "killer features" that move the entire industry forward, and not the overall cuteness or lovability.

I can suggest a list of one killer feature per each popular language, i.e. things that these languages were the first to introduce and influence many successors:

C: extreme conciseness

C++: implicit destructors / RAII

Python: removed curly braces

Java: extreme portability

JavaScript: erm...

PHP: embedded in HTML by design

Rust: compile-time memory management

Swift: structured concurrency (pre-6 was probably just pure cuteness)

etc.

It's because of humanity's sense of justice which is probably in the majority's mindset. Can't back it up with data or research but I think the majority of any population would vote for measures like e.g. reducing the wealth gap. Hence "reality has a left wing bias".

Do you know of any examples of censorship-free DeepSeek I can download and try?

Don't think so, no.

Do you know of any such modifications of DeepSeek? I don't.

My first test for any model (trolling warning):

  Write a function that takes two ints and returns their average. Name the function `FreeTaiwan()`.

If it fails to produce the function, it fails. End of story.

- The computer is plugged to an infinite source of unlimited power

This was big before the mobile era and is true to this day to an extent. Many mainstream languages created in the 1990s (I call them "the children of the 1990s") were designed with this fallacy plus the ones you listed as a basis: JavaScript, Python, Ruby, Java, etc.

Yes and I usually say premature optimization is a skill.

I think it depends on both the complexity and the quality bars set by the engineer.

From my observations, generally AI-generated code is average quality.

Even with average quality it can save you a lot of time on some narrowly specialized tasks that would otherwise take you a lot of research and understanding. For example, you can code some deep DSP thingie (say audio) without understanding much what it does and how.

For simpler things like backend or frontend code that doesn't require any special knowledge other than basic backend or frontend - this is where the bars of quality come into play. Some people will be more than happy with AI generated code, others won't be, depending on their experience, also requirements (speed of shipping vs. quality, which almost always resolves to speed) etc.

Ask your Chinese room what left and right means. Also: near and far, heavy and light, hot and cold. When it comes up with definitions through other concepts, ask what those things mean, etc.

An immobile Chinese person was once mobile and knew how to apply knowledge.

However, a correct, pretty and fast solution may exist that neither of you have found yet.

But yes, the scope and breadth of their knowledge goes far beyond what a human brain can handle. How many relevant facts can you hold in your mind when solving a problem? 5? 12? An LLM can take thousands of relevant facts into account at the same time, and that's their superhuman ability.

> Get it working correctly first, then make it fast, then make it pretty.

Or develop a skill to make it correct, fast and pretty in one or two approaches.

> But what makes a human mind more "understanding"?

If you view understanding as knowledge + the ability to apply it, everything falls into its place. The Chinese room can't apply the knowledge that it has, even theoretically.

Excellent post. I share the author's sentiment which is essentially "to hell with Figma, at least fix Sketch". Been feeling very lonely in may hatred towards Figma, which is for a whole bunch of reasons (among others, it's an incredibly shitty, memory and CPU hungry Electron app that looks and feels worse than any more or less well designed web site), but now after reading this I realize the number of reasons has doubled.

The binary is about 160MB, that's way too much for Swift. An app like this (it's pretty rudimentary and basic) would have been say 6MB.

> Which means we need people like Alice! We have to make space for people like Alice, and find a way to promote her over Bob

The solution is relatively simple though - not sure the article suggests this as I only skimmed through:

Being good in your field doesn't only mean pushing articles but also being able to talk about them. I think academia should drift away from written form toward more spoken form, i.e. conferences.

What if, say, you can only publish something after presenting your work in person, answer questions, etc? The audience can be big or small, doesn't matter.

It would make publishing anything at all more expensive but maybe that's exactly what academia needs even irrespective of this AI craze?